EidolonDB

Appearance

Concepts ​

What Is EidolonDB? ​

EidolonDB is a memory system for AI agents. Instead of only storing embeddings, it extracts candidate memories, classifies them into tiers, deduplicates low-signal data, and runs lifecycle policies to evolve memory over time.

How It Differs from a Vector DB ​

A vector DB is primarily storage + similarity lookup. EidolonDB adds memory behavior:

  • extraction from raw unstructured input
  • tiered memory state (short_term, episodic, semantic)
  • lifecycle automation (promotion, distillation, expiration, archival)
  • hybrid ranking (semantic + recency + importance)

The Three Tiers ​

  • short_term: immediate, volatile context; decays quickly
  • episodic: medium-horizon events and decisions
  • semantic: durable long-term facts and distilled knowledge

Memory Pipeline ​

  1. Ingest raw input (POST /ingest)
  2. Extract candidate memories with LLM
  3. Classify candidates by memory tier
  4. Deduplicate against existing memory
  5. Store accepted memory (if autoStore: true)
  6. Apply lifecycle rules over time (POST /lifecycle/run)